app.py

"""
☕ Coffee Barista v42 — Gradio Space
Chat-only coffee expert, no tools needed.
"""
import gradio as gr
from llama_cpp import Llama

MODEL_URL = "https://huggingface.co/ynanxiu/qwen25-3b-coffee-master-gguf/resolve/main/qwen25_3b_coffee_master_v43_q4km.gguf"
MODEL_PATH = "qwen25_3b_coffee_master_v43_q4km.gguf"

SYSTEM_PROMPT = "你是一个专业咖啡师，拥有丰富的咖啡知识和吧台经验。用自然、亲切的中文回答，像在跟熟客聊天一样。回复简洁、有人情味，1-3句话。"

# Download model on startup
import os
if not os.path.exists(MODEL_PATH):
    from huggingface_hub import hf_hub_download
    hf_hub_download(repo_id="ynanxiu/qwen25-3b-coffee-master-gguf", filename="qwen25_3b_coffee_master_v42_q4km.gguf", local_dir=".")

llm = Llama(model_path=MODEL_PATH, n_ctx=2048, n_threads=4, verbose=False)

def chat(message, history):
    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
    for h in history:
        messages.append({"role": "user", "content": h[0]})
        if h[1]:
            messages.append({"role": "assistant", "content": h[1]})
    messages.append({"role": "user", "content": message})
    
    response = llm.create_chat_completion(
        messages=messages,
        temperature=0.7,
        max_tokens=100,                              # ← 硬截断，≈ 150 汉字
        repeat_penalty=1.15,                         # ← 防复读机
        stop=["<|im_end|>", "<|im_start|>", "\n\n"], # ← 自然段落处停
    )
    return response["choices"][0]["message"]["content"]

demo = gr.ChatInterface(
    fn=chat,
    title="☕ 咖啡大师 Chat",
    description="一个懂咖啡、会聊天的 AI 咖啡师。问我任何咖啡问题！",
    examples=[
        "手冲咖啡水温怎么选？",
        "拿铁和卡布奇诺有什么区别？",
        "新手学咖啡从哪开始？",
        "深烘豆手冲有什么技巧？",
    ],
    theme="soft",
)

if __name__ == "__main__":
    demo.launch()